Enhanced Data Management

Enhanced Data Analysis for Environmental Observatory 

AI and DL technologies are increasingly utilized to analyze vast amounts of environmental data. These tools can uncover patterns and trends that are not easily detectable through traditional methods. For instance, motion-sensing cameras combined with AI can automate the identification of biodiversity, achieving high accuracy rates in species recognition. This capability is crucial for monitoring ecosystems and assessing the impact of human activities on wildlife. 

Generative AI, in the context of multimodal visual modeling, has emerged as a powerful approach for processing heterogeneous data. It can create new content based on the input it receives enhancing content creation across various fields, including various accessibility technologies. Supporting this, the integration of LLMs with linked-data technologies enriches the context available for decision-making by leveraging various data types, such as text, images, audio, and video across different domains. Multimodal generative AI can produce complex outputs that blend different media forms, enhancing user engagement and experience. In another point of view, if one modality is unreliable or absent, multimodal systems can still function effectively by relying on other available data sources. This resilience is crucial in real-world applications where data may be incomplete or noisy. Thus, multimodal visual modeling enhances understanding by integrating multiple data types leading to accurate interpretations and responses in applications like virtual assistants supporting better-informed decisions with comprehensive insights.

Generally, LLMs are inefficient with real-time sensor data, requiring frequent retraining. Linked data and semantic web technologies offer solutions by enabling dynamic, real-time querying through SPARQL, connecting various data sources. Only a few research works have been done using LLMs to generate SPARQL. Therefore, currently there is lack of methods to integrate and process information from diverse sources. The novelty of this research work lies in its potential to create a unified and collaborative data source for forest observatories, acting as a generative AI platform. We will leverage LLMs to automatically generate SPARQL queries that access sensor data. Instead of requiring continuous retraining on live data, LLMs will interact with the linked data framework, enabling dynamic, real- time querying. On top of this, we will develop AI-driven agent technologies that act as intelligent representatives for forest environments for better decision-making. Also, the proposed efficient DL methods and XAI techniques lead to improve model performance and trustworthiness, respectively. Thus, this integration would streamline information access and processing for authorities. 

Team: 

Ms. Charuni Kosala Jayaweera

Supervisors: Prof. Dulani Meedeniya